# Net vs Gross External Adjustment: Demographics as a Latent Factor

## Abstract

The headline finding of this paper is that demographics predict the income balance ($Z_1$=41.4***, p<0.001) but not the trade balance ($Z_1$=-3.0, NS): aging societies adjust externally through accumulated foreign asset returns, not trade competitiveness. This "income balance dominance" overturns the standard lifecycle narrative and reframes the global imbalances debate around yield channels rather than quantity channels. The income balance effect is rate-dependent: pre-GFC $Z_1$=55.6***, collapsing to 28.6 (NS) post-GFC as QE and the zero lower bound compressed the interest rate differentials through which demographics generate income flows (Chow F=46.3, p<0.001). Controlling for real bond yields absorbs the entire demographic-income balance signal (-123% attenuation), establishing a causal chain from demographics through rates to income flows. A second surprise: the savings-investment gap *suppresses* rather than mediates the demographic-CA link — controlling for S-I amplifies the $Z_1$ coefficient from 3.0 (NS) to 45.5, implying that demographic forces on the current account are entirely masked by offsetting savings-investment dynamics. Gross positions respond to demographics primarily through FX reserves ($Z_1$=63-67***), the only consistently significant instrument across specifications, while other gross position channels are fragile and specification-dependent. Shapley-Owen decomposition attributes 58% of CA variance to savings-investment channels, 21% to demographics directly, and 21% to gross positions — a more balanced partition than previously thought. Using data for up to 164 countries over 1970-2024, we conclude that demographics shape external balances primarily through the return structure of accumulated foreign positions, with the income balance as the dominant transmission channel.

## Introduction

A large literature documents that demographic structure — principally the age distribution summarized by old-age and youth dependency ratios — predicts current account balances across countries and over time. The standard interpretation follows lifecycle theory: aging societies save more than they invest domestically, generating current account surpluses. This net savings-investment view underpins the External Balance Assessment framework at the IMF and numerous academic studies.

A parallel literature examines bilateral capital flows using gravity models, finding that demographic differences between country pairs predict portfolio debt allocation but not foreign direct investment. These bilateral results point to a gross reallocation channel: aging societies do not merely save more in aggregate but actively reallocate portfolios toward foreign debt instruments in search of yield.

Three puzzles emerge from juxtaposing these literatures. First, demographics strongly predict bilateral portfolio debt flows but not FDI — the "FDI null." Second, in multilateral regressions, the entire current account effect operates through the income balance rather than the trade balance — "income balance dominance." Third, capital account openness amplifies the demographic effect in bilateral regressions but dampens it in multilateral CA regressions — the "KAOPEN asymmetry."

We investigate these puzzles through a unified gross/net decomposition. Our results partially support and partially challenge the initial hypothesis. Income balance dominance is strongly confirmed ($Z_1$=41.4*** on income balance, essentially zero on trade balance). However, the KAOPEN sign-flip proves fragile in the expanded sample: the $Z_1 \times$KAOPEN interaction on CA is positive but insignificant (+5.4, NS), while the income balance interaction is only marginally significant (-14.2*). The gross position channel is weaker still — significant only for FX reserves across all specifications — and the mediation analysis reveals a dramatic suppression pattern rather than the anticipated mediation, with the demographic CA coefficient amplifying fifteen-fold when the savings-investment gap is controlled.

### Related Literature

This paper bridges three strands of literature. First, the lifecycle savings approach to external balances, from Higgins (1998) and Chinn and Prasad (2003) through the IMF's EBA methodology. Second, the gravity model of bilateral portfolio allocation, including Lane and Milesi-Ferretti (2008) and our companion bilateral analysis. Third, the growing literature on gross capital flows and external balance sheets, initiated by Lane and Milesi-Ferretti (2007) and extended by Broner et al. (2013) and Forbes and Warnock (2012).

Our contribution is to show that these literatures address different manifestations of a single demographic latent factor — but that the channels interact in unexpected ways, with the savings-investment gap acting as a suppressor variable rather than a mediator.

## Data and Variables

### Panel Construction

We merge three data sources. The primary panel covers up to 164 countries over 1970-2024, drawn from the multilateral analysis: IMF Balance of Payments, International Investment Position statistics, World Bank WDI, and the Chinn-Ito KAOPEN index. We augment this with trade and income balance decompositions from the CA decomposition analysis (Peterson, 2026g, "The Demographic Erosion of Fiscal Leverage"), and reporter-year aggregated bilateral portfolio positions from the gravity bilateral analysis (normalized by reporter GDP).

### Demographic Variables

We use three principal components ($Z_1$, $Z_2$, $Z_3$) of the population age distribution, following the multilateral specification. $Z_1$ loads primarily on old-age dependency and captures the dominant demographic channel. We also employ 5-year lagged demographics and the predetermined old-age dependency ratio (20-year forward projection) for robustness.

### Gross Position Variables

From the IIP: gross external assets and liabilities (scaled by GDP), gross international financial integration (assets + liabilities), and the net foreign asset position. Instrument-level: FDI assets and liabilities, portfolio equity assets, debt assets and liabilities, and foreign exchange reserves. We construct net positions by instrument and year-on-year changes as flow proxies. Given the extreme skewness of gross positions (driven by financial centers like Luxembourg and Ireland), we report both raw and winsorized (p1/p99) results.

### Controls

Following the EBA methodology: fiscal balance (% GDP), lagged NFA (% GDP), log relative output per worker, and capital account openness (KAOPEN). All regressions employ Panel GLS with AR(1) error correction.

## Net Decomposition

### Current Account Components

We first replicate the current account decomposition. Regressing each CA component on demographic structure:

$$CA_{it} = \alpha_i + \gamma_t + \beta Z_{it} + \delta X_{it} + \varepsilon_{it}$$

The key finding confirms income balance dominance: $Z_1$=41.4*** (p<0.001) on the income balance, while the trade balance coefficient is essentially zero ($Z_1$=-3.0, NS). The full CA effect ($Z_1$=3.0, NS) is far smaller than the income balance effect — indeed, the headline CA coefficient is not statistically significant, because the trade balance adds noise that washes out the income balance signal. Pre-GFC, the income balance effect is even stronger ($Z_1$=55.6***), collapsing post-GFC ($Z_1$=28.6, NS). A Chow test confirms the structural break (F=46.3, p<0.001).

### The Suppression Puzzle

The savings-investment decomposition reveals a surprising pattern. Demographics predict national savings ($Z_1$=58.2***) and investment ($Z_1$=36.3***) — but not their gap ($Z_1$=-12.8, NS). Both savings and investment rise with aging, leaving the net effect ambiguous. More strikingly, when we control for the S-I gap in the CA regression (Baron-Kenny step 3), the demographic coefficient *increases* from 3.0 (NS) to 45.5 — a -1,414% "attenuation" indicating massive suppression. The CA coefficient amplifies fifteen-fold and shifts from insignificance to high significance when the S-I gap is absorbed.

This means the S-I gap almost entirely masks the demographic-CA relationship rather than transmitting it. The S-I gap captures business-cycle and policy variation that is correlated with demographics but works in the opposite direction of the structural demographic channel. Once this noise is absorbed, the true demographic effect on CA emerges at a magnitude comparable to the income balance coefficient (45.5 vs 41.4). This finding profoundly challenges the standard lifecycle interpretation where demographics operate through the S-I gap and instead suggests a more direct structural channel — plausibly through valuation effects, income flows, and portfolio composition rather than contemporaneous savings behavior. The suppression is far more dramatic than previously appreciated: without controlling for S-I, the demographic-CA link is essentially invisible.

## Gross Positions

### Aggregate Balance Sheet

Regressing gross external positions on demographic structure yields largely insignificant results in the full sample, with negative R-squared values indicating that the model fits poorly when financial center outliers dominate. Neither winsorizing nor excluding financial centers resolves the aggregate insignificance:

**Winsorizing** gross positions at p1/p99 does not produce significant aggregate results: gross liabilities ($Z_1$=124, NS), gross IFI ($Z_1$=57, NS). The aggregate demographic signal on gross positions that appeared in the 69-country panel does not survive expansion to the 140-country sample.

**Excluding financial centers** (LUX, IRL, HKG, SGP, CHE, NLD, BEL) similarly fails to clarify the aggregate picture: gross IFI remains insignificant ($Z_1$=321, p=0.597), as do individual aggregate measures.

### Instrument Decomposition

Among instruments, only one stands out consistently. **FX reserves** show the strongest and most robust demographic signal: $Z_1$=66.7*** (full sample), 63.0*** (excl FC), 66.4*** (winsorized). Aging economies accumulate reserves — potentially reflecting precautionary demand in aging societies with large external positions, or the export-led growth strategies of aging East Asian economies. FX reserves are the sole instrument-level result that survives all sample treatments.

**Debt liabilities** do not achieve significance in any specification: $Z_1$=63.7 (NS, full sample), 123.6 (NS, excl FC), 63.6 (NS, winsorized). The debt liability channel that appeared in the smaller panel does not survive the expanded sample, weakening the liability-side story.

Notably, **portfolio equity assets** show an unexpected negative significance ($Z_1$=-184.2*, full sample; -131.8**, winsorized), suggesting that aging societies actually *reduce* equity exposure abroad — the opposite of the lifecycle diversification prediction. Debt *assets* and FDI remain insignificant in the multilateral IIP data, even though bilateral gravity regressions find strong effects for portfolio debt. This discrepancy likely reflects measurement: bilateral CPIS data captures granular portfolio allocation, while IIP debt_assets aggregates all external debt claims including bank deposits and official lending.

### Lifecycle vs Hot Money

The fragility of the aggregate gross signal raises a legitimate concern about whether any demographic pattern in gross data reflects lifecycle savings or financial center noise. Table 4d tests this by decomposing positions into lifecycle instruments (FX reserves + debt liabilities + FDI liabilities — stable, structural positions) and speculative instruments (portfolio equity — volatile, short-horizon). The lifecycle composite is insignificant ($Z_1$=222, NS, full sample; 287, NS, excl FC), while portfolio equity is marginally significant but negative ($Z_1$=-184*, full sample). The lifecycle/hot-money decomposition thus does not cleanly separate the channels — the negative portfolio equity coefficient may reflect aging societies' shift from equity to fixed-income instruments, consistent with lifecycle theory but operating through asset composition rather than aggregate volume.

Iterative trimming of gross IFI provides modest support for an underlying demographic signal. Progressively trimming at p5/p95 and p10/p90: $Z_1$=54 (NS, raw), 128 (NS, p5/p95), 217* (p=0.085, p10/p90). On gross liabilities, trimming is more effective: $Z_1$=123 (NS, raw), 162* (p=0.075, p5/p95), 166** (p=0.022, p10/p90). The demographic signal on gross liabilities strengthens monotonically with outlier removal, eventually reaching conventional significance at p10/p90 — but requires aggressive trimming that drops substantial sample mass.

## KAOPEN Gating

### The Sign-Flip: Suggestive but Fragile

The KAOPEN interaction shows the hypothesized sign pattern but with diminished statistical power in the expanded sample:

- **Current account**: $Z_1 \times$KAOPEN = +5.4 (NS, full), +6.3 (NS, excl FC). The sign is positive — openness is associated with a larger demographic CA surplus — but the interaction is not statistically significant.
- **Income balance**: $Z_1 \times$KAOPEN = -14.2* (full, p<0.10), -14.1 (NS, excl FC). Openness is marginally associated with dampening the income balance effect, but significance is fragile and lost when financial centers are excluded.

The gross position interactions (assets, liabilities, IFI) remain uniformly insignificant, confirming that KAOPEN does not gate the *stock* of gross positions. The direction of effects is consistent with a return-channel story — openness potentially allowing income flows to escape more freely — but the expanded sample does not provide the statistical power to confirm this mechanism. The sign pattern is suggestive rather than conclusive.

The bilateral gravity paper's finding that $Z \times$KAOPEN strongly amplifies capital *flows* between country pairs thus does not translate into a robust multilateral KAOPEN interaction on either the CA or the income balance. This may reflect that the bilateral identifying variation (between-pair differences) is more powerful than the within-country panel variation available in multilateral regressions, or that the KAOPEN gating mechanism operates at the bilateral margin without aggregating to a detectable multilateral signal.

The Chow test confirms a structural break in the income balance channel at the GFC (F=46.3, p<0.001), with $Z_1$ collapsing from 55.6*** pre-GFC to 28.6 (NS) post-GFC. This parallels the monetary policy paper's central finding that QE and the ZLB masked demographic effects on interest rates. The mechanism is direct: the income yields on demographically-accumulated foreign positions converged toward zero during the QE era. The "masking" of demographic signals documented in the monetary paper thus propagates into the external balance through the income account — QE simultaneously obscured the demographic-rate relationship and the demographic-income balance relationship, because the latter operates through the former. As the monetary paper documents re-emergence of the demographic-rate signal in 2022--2024 post-QE tightening, we should expect corresponding re-emergence of the income balance channel as rate differentials widen.

## Bilateral Aggregation Bridge

We aggregate bilateral portfolio positions (from CPIS/CDIS) by reporter-year, normalize by reporter GDP, and compare coefficients on the intersection sample (1,186 obs, 75 countries). Bilateral aggregated positions and IIP equivalents are both largely insignificant: bilateral portfolio total $Z_1$=527 (NS) vs IIP gross assets $Z_1$=1,491 (NS); bilateral portfolio debt $Z_1$=-4.4 (NS) vs IIP debt assets $Z_1$=676 (NS). The bilateral FDI aggregate is similarly null ($Z_1$=-4.9, NS) as is IIP FDI assets ($Z_1$=1,493, NS). The FDI null persists at both levels, and the aggregate IIP debt assets signal that appeared significant in the smaller panel is now NS.

The power gap is unsurprising: bilateral CPIS coverage is limited, and the aggregation discards the bilateral variation that gives gravity models their identifying power. Partner-count heterogeneity does not reveal a clear pattern: neither the many-partner subsample ($Z_1$=286, NS on bilateral total; 359, NS on IIP assets) nor the few-partner subsample ($Z_1$=24, NS on bilateral total; 1,995, NS on IIP assets) yields significant results, though the few-partner subsample shows a significant CA coefficient ($Z_1$=86.7***).

### Reconciling Bilateral Strength with Aggregate Noise

A natural question is how the gravity bilateral paper's strong portfolio debt signal ($\Delta Z_1$=3.88***, joint Wald $\chi^2$=108.6) coexists with this paper's finding that aggregate gross positions are "noisy." Table 8d addresses this directly by re-estimating bilateral-aggregated and IIP positions excluding financial centers. The bilateral portfolio debt coefficient improves from $Z_1$=-4.4 (NS) to $Z_1$=208.6* (p<0.10) — a shift from complete insignificance to marginal significance. IIP debt assets similarly strengthen from $Z_1$=676 (NS) to $Z_1$=1,118** (p<0.05).

Excluding financial centers thus partially recovers the bilateral signal, but the magnitudes remain imprecise and the bilateral-aggregated result only reaches marginal significance. The CPIS bilateral matrix captures portfolio allocation between end-investor countries; positions routed through Luxembourg, Ireland, or Singapore are recorded at the bilateral level with their ultimate counterpart, while IIP aggregates include the intermediary's entire balance sheet. When financial centers are excluded, the aggregate gross signal moves toward significance, partially validating the financial center noise hypothesis — but the improvement is less dramatic than in the smaller panel, suggesting that the expanded country coverage introduces additional heterogeneity beyond financial center intermediation.

## Variance Decomposition

### Baron-Kenny Mediation

The mediation analysis yields three key findings:

**Path A (S-I gap)**: Massive suppression, not mediation. Adding the S-I gap to the CA regression amplifies the Z₁ coefficient from 3.0 (NS) to 45.5 — a -1,414% "attenuation" indicating that the S-I gap almost entirely masks the demographic-CA relationship. The S-I gap is a *suppressor variable* — correlated with both demographics and CA but capturing different (opposing) variation. This dramatically overturns the standard interpretation that demographics affect CA through the savings-investment channel. The suppression is far more extreme than previously appreciated: without controlling for S-I, the demographic-CA link is statistically invisible.

**Path B (Gross IFI → Income balance)**: Negligible mediation. Controlling for gross IFI attenuates the income balance Z₁ coefficient by only 1.4%. Gross financial integration does not meaningfully mediate the demographic-income balance relationship — the income effect operates through composition and returns, not volume.

**Path C (Debt assets → Income balance)**: Small attenuation (4.1%). Debt asset positions provide marginal mediation of the income balance effect, but the overwhelming majority of the demographic signal operates through channels other than gross debt positions.

### Shapley-Owen Decomposition

The Shapley decomposition of CA R-squared across three variable groups (common sample: 3,834 obs, 149 countries) reveals a more balanced partition than anticipated: savings-investment variables account for 57.7% of explained variance, demographics directly for 21.4%, and gross position variables for 20.8%. While the S-I channel remains the single largest contributor, the demographic and gross position shares are substantially larger than in the smaller panel — demographics directly account for over a fifth of CA variance, and gross positions contribute an equal share. The total R-squared is modest (0.074), reflecting the well-known difficulty of explaining current account variation.

## Dynamics

### First Differences and Levels

First-differenced gross positions (year-on-year changes) show no demographic signal — all Z coefficients are NS with near-zero R-squared. Demographics shape the *level* of external positions, not annual *flows*. This is consistent with demographics as a slow-moving structural force that establishes long-run portfolio equilibria rather than driving short-term capital movements.

### Lagged Demographics

Five-year lagged demographics do not significantly predict the income balance ($Z_1$=19.3, NS), CA ($Z_1$=-17.2, NS), gross assets ($Z_1$=-67.1, NS), or debt assets ($Z_1$=-11.1, NS). The contemporaneous Z is also not significant on CA (3.0, NS). The lagged result that appeared in the smaller panel does not survive the expanded sample, weakening the evidence for slow-moving demographic effects operating through accumulated past positions. This is consistent with the broader pattern of attenuation in the expanded panel.

### Structural Break

A formal Chow test confirms a structural break at the GFC for the income balance (F=46.3, p<0.001). Pre-GFC: $Z_1$=55.6***. Post-GFC: $Z_1$=28.6 NS. This parallels the structural break documented in the monetary policy analysis for the demographic-rate relationship. The income balance channel appears rate-dependent: when interest rates are higher (pre-GFC), the income flow consequences of demographically-driven portfolio positions are larger. Near-zero rates compress income differentials, muting the channel.

Gross positions themselves do not exhibit a clear structural break — pre- and post-GFC coefficients are both insignificant. The break operates through *returns on positions*, not *positions themselves*.

### Linking to the Monetary Channel

The rate-dependence of the income balance channel suggests a direct link to the monetary policy paper's finding that QE and the zero lower bound masked demographic effects on interest rates. Table 13c tests this by adding real bond yields as a control to the income balance specification. The result is striking: pre-GFC, controlling for real_bond_10y reduces the income balance $Z_1$ coefficient from 55.6*** to -13.0 (NS) — a -123% attenuation. Rates absorb virtually the entire demographic-income balance signal. Post-GFC, where the baseline $Z_1$ is already weak (28.6, NS), adding rates reduces it further (0.9, NS), though the post-GFC sample with rate data is very small (N=230).

This establishes a clear causal chain: demographics → interest rate differentials → portfolio positions → income flows → current account. When QE compressed rate differentials toward zero, the income yields on demographically-accumulated foreign positions converged, muting the income balance channel. The structural break in the demographic-income balance relationship at the GFC is not a breakdown of the demographic mechanism but a predictable consequence of rate compression documented in the monetary paper. As rates normalize post-QE, this channel should re-emerge — a testable prediction consistent with the monetary paper's finding of demographic re-emergence in 2022-2024 rate data. Note that the rate-mediated analysis is necessarily limited to the OECD subsample with bond yield data (N=711 full, N=435 pre-GFC, N=230 post-GFC), so these results should be interpreted with the caveat of restricted country coverage.

## Robustness

The robustness matrix across 5 dependent variables and 9 specifications reveals that the income balance result, while the most robust individual finding, is more fragile than in the smaller panel. It is significant in the full sample (41.4***), non-OECD (59.8***), excluding financial centers (41.8***), low-income tercile (146.2***), and with predetermined demographics OADR+20 (9.6***). However, it is now NS in the OECD subsample (-1.9), the middle-income tercile (-5.9), and notably in the high-income tercile (19.7, NS) — a result that was previously significant. The alternative demographics specification (old_dep, youth_dep) also yields an insignificant result (8.7, NS).

The OECD null on income balance is notable and extends to all dependent variables — the OECD subsample shows no detectable demographic signal on any external balance measure. This raises important questions about identification: the OECD subsample, which provides the cleanest data and most reliable measurement, shows no demographic-external balance relationship. The income balance result is driven entirely by non-OECD countries, particularly the low-income tercile where demographic variation is largest but measurement quality is weakest.

The CA result is weaker still: the full sample coefficient ($Z_1$=3.0) is not significant. Only the high-income tercile (39.1**) yields a significant CA coefficient. The alternative demographic specifications are also NS (alt demo: -8.0; OADR+20: -1.9). The headline CA result is essentially null in the expanded panel.

Gross position results are fragile and specification-dependent, as expected given the financial center noise and measurement issues discussed above. The OECD subsample shows negative and significant gross asset and debt asset coefficients ($Z_1$=-1,285** and -596**, respectively), suggesting that within the OECD, aging is associated with *smaller* external balance sheets — the opposite of the lifecycle prediction.

## Conclusion

The central finding of this paper is income balance dominance: demographics predict the income balance ($Z_1$=41.4***) but not the trade balance ($Z_1$=-3.0, NS), and indeed not even the headline current account ($Z_1$=3.0, NS). Aging societies adjust externally not through trade competitiveness but through the yield on accumulated foreign assets. This is the "how" behind the current account effects documented in the global imbalances literature, and it reframes the demographic channel from a quantity story (more saving, less investment) to a return story (portfolio accumulation generating income flows).

Three supporting findings reinforce and extend this conclusion, though with important caveats about the fragility of some channels in the expanded 140-country panel.

First, the savings-investment gap is a *suppressor*, not a *mediator*, of the demographic-CA relationship. Controlling for S-I amplifies the demographic coefficient fifteen-fold, from 3.0 (NS) to 45.5 (-1,414% attenuation), implying that demographic forces on the current account are almost entirely masked by offsetting savings-investment dynamics. Without the S-I control, the demographic-CA link is statistically invisible. For IMF External Balance Assessment (EBA) methodology, this means that demographic "norms" estimated through the S-I channel systematically understate the true demographic undertow in current accounts. Surveillance frameworks should treat the income balance coefficient, not the CA coefficient, as the relevant magnitude of demographic pressure.

Second, the income balance channel is rate-dependent, collapsing after the GFC (Chow F=46.3, p<0.001) when QE and the ZLB compressed interest rate differentials. Controlling for real bond yields absorbs the entire demographic-income balance signal pre-GFC (-123% attenuation), establishing a direct link to the monetary paper's finding that unconventional policy masked demographic effects on rates. The masking is thus not confined to interest rate regressions — it propagates through the income account into external balance dynamics. As the monetary paper documents re-emergence of the demographic-rate signal in 2022--2024, we predict corresponding strengthening of the income balance channel as rate differentials normalize.

Third, gross external positions respond to demographics almost exclusively through FX reserves ($Z_1$=63-67***), the only consistently significant instrument across all specifications. Other gross position channels — including debt liabilities and the aggregate balance sheet — do not survive the expanded panel. The bilateral gravity signal is strong precisely because it captures bilateral variation unavailable in multilateral regressions; when financial centers are excluded, bilateral-aggregated portfolio debt improves from $Z_1$=-4.4 (NS) to $Z_1$=209* (p<0.10), and IIP debt assets strengthen to $Z_1$=1,118** — but the bilateral aggregate remains only marginally significant. The Shapley decomposition shows a more balanced partition than previously thought: savings-investment channels explain 58% of CA variance, demographics directly 21%, and gross positions 21%.

Fourth, our findings reconcile an apparent contradiction across companion papers. The bilateral gravity analysis finds that financial openness *amplifies* demographically driven capital flows (all $Z \times$KAOPEN p < 0.023), while the causal identification paper finds that capital account opening *weakens* the demographic-CA channel in transition economies (9:1 pre-to-post coefficient ratio). The resolution operates through three layers. The gravity paper measures bilateral gross *positions* — openness enables more cross-border holdings to respond to demographic distance, a volume effect. This paper shows that the net CA effect of demographics operates through income balances, not trade or gross positions, and that KAOPEN gates *returns* rather than *stocks* (gross position interactions all p > 0.48). The causal paper demonstrates that the pre-opening $Z_1$ coefficient was inflated by structural confounders — Soviet-era institutional characteristics that correlate with demographics but are disrupted by opening (partial $R^2$ of $Z_1$ on observables drops from 0.65 to 0.06 post-opening). Thus: openness amplifies bilateral gross flows driven by demographic distance (gravity); the net CA effect channels through income balances that openness compresses via return convergence (this paper); and the cross-sectional $Z_1$ → CA relationship captures institutional configurations that opening disrupts rather than a lifecycle savings pipeline that opening activates (causal identification). The income balance is the missing link — it explains how demographics can strongly predict bilateral portfolio allocation without proportionally moving the net current account position.

An important caveat is that the OECD subsample shows no detectable demographic signal on any external balance measure, including the income balance. The income balance dominance result is driven by non-OECD countries, particularly the low-income tercile. This raises questions about whether the finding reflects a genuine structural mechanism or is influenced by the greater demographic variation but weaker measurement quality in developing economies.

These findings reframe the demographic external adjustment story around the return structure of external positions rather than gross position volumes. The practical implication is that aging societies' external adjustment is more rate-sensitive than standard models suggest — and that the QE era temporarily obscured this reality by compressing the very yield differentials through which demographics generate external adjustment. However, the fragility of several channels in the expanded panel cautions against overconfidence in the precise magnitudes.

## References

Baron, R.M. and Kenny, D.A. (1986). The moderator-mediator variable distinction in social psychological research. *Journal of Personality and Social Psychology*, 51(6), 1173-1182.

Broner, F., Didier, T., Erni, A. and Schmukler, S.L. (2013). Gross capital flows: dynamics and crises. *Journal of Monetary Economics*, 60(1), 70-94.

Chinn, M.D. and Ito, H. (2006). What matters for financial development? Capital controls, institutions, and interactions. *Journal of Development Economics*, 81(1), 163-192.

Chinn, M.D. and Prasad, E.S. (2003). Medium-term determinants of current accounts in industrial and developing countries. *Journal of International Economics*, 59(1), 47-76.

Chow, G.C. (1960). Tests of equality between sets of coefficients in two linear regressions. *Econometrica*, 28(3), 591-605.

Forbes, K.J. and Warnock, F.E. (2012). Capital flow waves: surges, stops, flight, and retrenchment. *Journal of International Economics*, 88(2), 235-251.

Higgins, M. (1998). Demography, national savings, and international capital flows. *International Economic Review*, 39(2), 343-369.

Lane, P.R. and Milesi-Ferretti, G.M. (2007). The external wealth of nations mark II. *Journal of International Economics*, 73(2), 223-250.

Lane, P.R. and Milesi-Ferretti, G.M. (2008). International investment patterns. *Review of Economics and Statistics*, 90(3), 538-549.

Phillips, S., Catao, L., Ricci, L., Bems, R., Das, M., Di Giovanni, J., Unsal, D.F., Castillo, M., Lee, J., Rodriguez, J. and Vargas, M. (2013). The External Balance Assessment (EBA) methodology. *IMF Working Paper* 13/272.
